Network tool review, week 2026-W43: 17 eligible services, 288 tools sampled

What the recorded evidence covers

Observation week (UTC): 2026-10-19 – 2026-10-25. Published 2026-10-26. The observation period is a completed ISO week; the publication date is when this analysis first became available.

There are 1016 visible, usable tool tests, comprising 1016 imported from GreatFire Circumvention Central and 0 submitted by BreakHub visitors. These are test records, not distinct people. Direct-connection (“None”), hidden and junk-name rows are excluded; spelling aliases are grouped at query time. Across 288 sampled tools, 36 meet the minimum of 5 tests spanning 48 hours. The largest tool accounts for 16.0% of these records. Coverage spans 7 UTC days, 4 reported location categories and 5 recorded operating systems; a category count does not imply balanced coverage.

The catalogue identifies 17 eligible services. The comparison table uses services only, which avoids treating a configurable client as if it were one provider. Clearing the minimum makes an entry eligible for a descriptive ranking; it does not establish a reliable recommendation.

Up to five eligible entries

The order uses the site's existing algorithm: medians per tool, competition ranks for speed (higher), latency (lower) and stability (higher), then the mean of those three ranks (lower score first). Missing metrics receive pool size + 1; score ties are ordered by speed, test count and name. This is a one-week pool across all reported locations and systems, so it need not match the default 60-day rankings. Speed is converted from bytes per second to decimal Mbps. “—” means a missing metric, not zero. Span measures first to last test, not continuous monitoring.

Dispersion behind the medians

P25–P75 contains the middle half of recorded values; n counts non-missing readings for that metric. A narrow middle range can still hide failures in the tails. A broad range means the median alone leaves out substantial variation. These ranges are descriptive, not confidence intervals; with only a few readings they are especially fragile. 0 tests lack speed, 0 lack latency and 0 lack stability. Each metric therefore may describe a different subset. Stability is the image-dimension pass rate in the test, not a measure of uptime, privacy or safety.

What can be compared with the previous week

The preceding week (2026-10-12 – 2026-10-18 UTC) has 1161 tests across 306 sampled tools; 36 meet the count/span floor, compared with 36 this week. Imported / visitor counts change from 1161 / 0 to 1016 / 0. The largest tool's share changes from 15.8% to 16.0%. These are changes in the collected sample, not estimates of adoption or market share.

For speed comparisons, each tool needs at least 10 finite speed readings in each week, the same pool membership rule, and a span of at least 48 hours in each week. The following 5 eligible comparisons show the two medians and IQR widths side by side. A wider IQR means a wider middle half of the recorded speeds; it does not by itself explain the cause.

Both weeks pool all reported locations and operating systems. The source, location, device, and tester mix can change between weeks; these are descriptive comparisons, not matched experiments or evidence that a provider caused a change. Repeated submissions may come from the same person. Check source and location filters before applying any apparent movement to your own connection.

How to use this evidence

1. Open a tool's detail page and inspect its counts, dates, region, operating system and source breakdown before considering its position. Prefer a short trial over a decision based on one rank.
2. Read How to read the rankings and Choosing a tool. When your own setting has few tests, the pooled table is background context, not a substitute for local evidence.
3. Run a test on the connection and device you actually use at different times on different days. Record the provider name consistently; client names alone can mix independent services. Repeated tests add observations, not independent users.
4. Inspect IP and connection information separately when exit-network properties matter. A high speed or stability value cannot establish identity protection; see How we measure for browser-test limits.

Provenance and scope

This article computes sample quality from stored GreatFire imports and BreakHub visitor results. It does not restate the observatory's outage, reachability or Tor digest and does not infer political causes. GreatFire attribution describes the source of imported records; this review is BreakHub's analysis, not a GreatFire endorsement. Import freshness was checked before first publication, but it cannot prove that every upstream record arrived. This completed-week article remains fixed after publication; newer field evidence appears in the guide supplements and the next eligible weekly article.